Publication detail

Deep convolutional neural network-based segmentation and classification of difficult to define metastatic spinal lesions in 3D CT data

CHMELÍK, J. JAKUBÍČEK, R. WALEK, P. JAN, J. OUŘEDNÍČEK, P. LAMBERT, L. AMADORI, E. GAVELLI, G.

Original Title

Deep convolutional neural network-based segmentation and classification of difficult to define metastatic spinal lesions in 3D CT data

English Title

Deep convolutional neural network-based segmentation and classification of difficult to define metastatic spinal lesions in 3D CT data

Type

journal article

Language

en

Original Abstract

This paper aims to address the segmentation and classiffcation of lytic and sclerotic metastatic lesions that are diffcult to deffne by using spinal 3D Computed Tomography (CT) images obtained from highly pathologically affected cases. As the lesions are ill-deffned and consequently it is diffcult to find relevant image features that would enable detection and classiffcation of lesions by classical methods of texture and shape analysis, the problem is solved by automatic feature extraction provided by a deep Convolutional Neural Network (CNN). Our main contributions are: (i) individual CNN architecture, and pre-processing steps that are dependent on a patient data and a scan protocol - it enables work with different types of CT scans; (ii) medial axis transform (MAT) post-processing for shape simpliffcation of segmented lesion candidates with Random Forest (RF) based meta-analysis; and (iii) usability of the proposed method on whole-spine CTs (cervical, thoracic, lumbar), which is not treated in other published methods (they work with thoracolumbar segments of spine only). Our proposed method has been tested on our own dataset annotated by two mutually independent radiologists and has been compared to other published methods. This work is part of the ongoing complex project dealing with spineanalysis and spine lesion longitudinal studies.

English abstract

This paper aims to address the segmentation and classiffcation of lytic and sclerotic metastatic lesions that are diffcult to deffne by using spinal 3D Computed Tomography (CT) images obtained from highly pathologically affected cases. As the lesions are ill-deffned and consequently it is diffcult to find relevant image features that would enable detection and classiffcation of lesions by classical methods of texture and shape analysis, the problem is solved by automatic feature extraction provided by a deep Convolutional Neural Network (CNN). Our main contributions are: (i) individual CNN architecture, and pre-processing steps that are dependent on a patient data and a scan protocol - it enables work with different types of CT scans; (ii) medial axis transform (MAT) post-processing for shape simpliffcation of segmented lesion candidates with Random Forest (RF) based meta-analysis; and (iii) usability of the proposed method on whole-spine CTs (cervical, thoracic, lumbar), which is not treated in other published methods (they work with thoracolumbar segments of spine only). Our proposed method has been tested on our own dataset annotated by two mutually independent radiologists and has been compared to other published methods. This work is part of the ongoing complex project dealing with spineanalysis and spine lesion longitudinal studies.

Keywords

CT analysis; spinal metastasis; convolutional neural network; computer aided detection

Released

01.10.2018

Publisher

Elsevier B.V.

Location

Amsterdam, The Netherlands

Pages from

76

Pages to

88

Pages count

13

URL

BibTex


@article{BUT149034,
  author="Jiří {Chmelík} and Roman {Jakubíček} and Petr {Walek} and Jiří {Jan} and Petr {Ouředníček} and Lukáš {Lambert} and Elena {Amadori} and Giampaolo {Gavelli}",
  title="Deep convolutional neural network-based segmentation and classification of difficult to define metastatic spinal lesions in 3D CT data",
  annote="This paper aims to address the segmentation and classiffcation of lytic and sclerotic metastatic lesions that are diffcult to deffne by using spinal 3D Computed Tomography (CT) images obtained from highly pathologically affected cases. As the lesions are ill-deffned and consequently it is diffcult to find relevant image features that would enable detection and classiffcation of lesions by classical methods of texture and shape analysis, the problem is solved by automatic feature extraction provided by a deep Convolutional Neural Network (CNN). Our main contributions are: (i) individual CNN architecture, and pre-processing steps that are dependent on a patient data and a scan protocol - it enables work with different types of CT scans; (ii) medial axis transform (MAT) post-processing for shape simpliffcation of segmented lesion candidates with Random Forest (RF) based meta-analysis; and (iii) usability of the proposed method on whole-spine CTs (cervical, thoracic, lumbar), which is not treated in other published methods (they work with thoracolumbar segments of spine only).

Our proposed method has been tested on our own dataset annotated by two mutually independent radiologists and has been compared to other published methods. This work is part of the ongoing complex project dealing with spineanalysis and spine lesion longitudinal studies.",
  address="Elsevier B.V.",
  chapter="149034",
  doi="10.1016/j.media.2018.07.008",
  howpublished="print",
  institution="Elsevier B.V.",
  number="C",
  volume="49",
  year="2018",
  month="october",
  pages="76--88",
  publisher="Elsevier B.V.",
  type="journal article"
}